Overclock.net banner

641 - 653 of 653 Posts

·
Facepalm
Joined
·
9,883 Posts
In attempting a shunt mod, I appear to have bridged the resistor to the an existing connection on the PCB board.

I'm assuming I'm going to need to clean this up, no clue where what this is connected to.
Why didn't you use kapton tape / high temp polyimide tape like I suggested several times in this thread?
Sure, the very experienced solderers may be able to get away with not using it, but this is absolutely something a novice should NOT skip!
I may get flamed for it but I'll keep suggesting people do it, even if the hotshots keep flaming me for it, until I get banned. Because I don't care about hotshot ultra popular people.

Use either this

or https://www.amazon.com/NTE-Electronics-SW02-10-No-Clean-Blue-098/dp/B0195UVWJ8/
to get the solder off.

And don't leave it there. Could be one of the 12v lines.
Then when you clean it all off, properly cover around the shunts with this.


Make sure you are using flux when applying shunts, and flux after every single step.
And why do you have so much solder on top of the shunt? You should be building a solder bridge inside the "FE gap" (the lower part of the original shunts, by the black housing), so when you put the new shunt on top (once again fluxing both before and after), you then melt the solder bridge between the original and new shunt, so then shunts are now flush. Then carefully siphon the excess solder off back to the iron.

It's a lot easier to solder shunts that dont have depressed metal edges (e.g. the 1W flush shunts on the Asus, MSI and eVGA cards), as then you don't need to even do that 'bridge' first.
 
  • Rep+
Reactions: Hueristic

·
Registered
Joined
·
10 Posts
Thanks for the info.

After trying the other resistor on the other side of the power connector (the one which is right by another component), I think I've come to the conclusion that I should not proceed with attempting to stack shunts on the existing resistors. I'm going to use my solder wick to remove the two I've attempted so far. I missed/forgot your recommendation for Klapton tape, but at this point I think I'm going to be happy if I can revert what I've done so far without damaging the card.

While I think this is something I could figure out, two of the six resistors are uncomfortably close to other components that even with tape I'm concerned about my ability to not mess something up.

I'll throw this under water and see what perf I get. I can also attempt the mod again in the future.
 

·
Facepalm
Joined
·
9,883 Posts
Thanks for the info.

After trying the other resistor on the other side of the power connector (the one which is right by another component), I think I've come to the conclusion that I should not proceed with attempting to stack shunts on the existing resistors. I'm going to use my solder wick to remove the two I've attempted so far. I missed/forgot your recommendation for Klapton tape, but at this point I think I'm going to be happy if I can revert what I've done so far without damaging the card.

While I think this is something I could figure out, two of the six resistors are uncomfortably close to other components that even with tape I'm concerned about my ability to not mess something up.

I'll throw this under water and see what perf I get. I can also attempt the mod again in the future.
It's fine to stack the shunts. This is exactly what I did. And it became much easier when I bought a decent (TS-100) soldering iron. MUCH easier.
The issue with the "excess solder" (not talking about your short, which MUST be removed), is for the BACKPLATE Side of the card. There is almost no clearance between a stacked shunt and the backplate on that side of the card, so it's very important to have the shunt fully flush and level, and not to have any excess solder sticking "up" 'from the top of the stacked shunt. A huge protip is to apply a small strip of that Polyimide 3M tape onto the backplate, right behind the shunt to make sure that there isn't any contact. Then screw the backplate on fully, all the torx screws all the way, then immediately unscrew and remove it, and check to make sure the tape isn't dented or ripped. If it's dented, it means either the shunt isn't fully flat, or you have excess solder on top :)

On the core side of the card, you won't have these clearance issues.

Anyway, you just need to know how to do the stacking. It's perfectly safe. It's a LOT harder to REMOVE (replace) the original shunts due to how much heat the PCB will absorb from the iron (you would need a >65W soldering iron and possibly a hot air station to heat the PCB enough for the iron to be able to melt the solder).

Again the key is to use the polyimide tape around the shunts, at least an inch around each one, that way you have plenty of space to work with. Then if you make a soldering accident, you're protected.

I wrote this post about a month ago in the other thread. Might be useful for you to look it over and make sure you're covered.
Notice that I have two videos linked. Oldirdey's video is best for "flush" shunts, which the 3090 FE (and the Gigabyte cards) do NOT have. But it is still helpful and educational to look at his video (start at 15:00).

The second video shows the "bridge" method. That card is an EVGA Card so the 'bridge' method is not required--oldirdey's method is best for those cards. But this is the method you need for the FE card. You also need to remember, when doing the 'bridge' method, that, after applying solder to the iron tip (and remembering to FLUX the metal contact area!), you MUST apply the iron to the edge of the shunt (preferably set to 380C), long enough so that the shunt heats up enough so the solder will "flow" to the lower metal connector and stay attached to it. It usually takes about 5-10 seconds of contact +moving the iron tip back and forth, for the shunt to heat up enough for the solder to flow. The edge shunts (e.g. 8 pins) will take less time because there's less copper around that part of the board to eat up all the heat. The PCIE Slot, GPU Chip Power and SRC shunts have a lot more stuff going on around them, so that will take more time.

Remember--kapton tape is your friend here. I highly recommend the 3M polyimide tape I both linked and suggested. Just make sure that the tape isn't 'blocking' the edge of the shunt (you'll have enough to easily cut/trim and get a nice good application). It also helps to clean the PCB with 91%-99% alcohol before applying any tape.

Also as I mentioned in the other thread, it pays HUGE dividends to have ANY spare old motherboard or video card with pre-existing solder joints (properly sized for applying shunts) to use as a practice board, which you can use to first solder any random shunt to that board, and learn how to deal with the PCB eating up most of the heat, and then stacking a shunt on top of your first shunt :) Then just desolder it. Once you can do that successfully, you're in excellent shape.

 

·
Registered
Joined
·
10 Posts
Great tips, thanks. I'll make sure to save this for later.

I think it doesn't help I bought a mediocre solder kit and me being a novice at soldering had oxidized my super skinny tip (which had terrible heat transfer at the tip but that's another issue) so I was attempting with a small but not tiny tip.

I think the extra motherboard is a great idea if I try this again. I've also got some extra graphics cards so I should definitely play around on those before trying this again.

Assuming I can revert the 2 resistors I soldered, clean up the short, and get everything working again, it may actually be better for me to first check performance of my card under the water block if nothing else as a baseline. I'm able to get 2070-2100 in games from this card before it starts to limit thermally. This feels like a card that could really shine with more power (and under water) but obviously that won't matter if I brick it because of my ****ty soldering skills and/or not being prepared enough.

Regarding clearance issues: I've got the EK 3090 FE waterblock so I may have different issues (and it's possible that stacking would never work with this block so in hindsight I should have confirmed that first). I'll check on that when do I my water block assembly later this afternoon.
 

·
Facepalm
Joined
·
9,883 Posts
Great tips, thanks. I'll make sure to save this for later.

I think it doesn't help I bought a mediocre solder kit and me being a novice at soldering had oxidized my super skinny tip (which had terrible heat transfer at the tip but that's another issue) so I was attempting with a small but not tiny tip.

I think the extra motherboard is a great idea if I try this again. I've also got some extra graphics cards so I should definitely play around on those before trying this again.

Assuming I can revert the 2 resistors I soldered, clean up the short, and get everything working again, it may actually be better for me to first check performance of my card under the water block if nothing else as a baseline. I'm able to get 2070-2100 in games from this card before it starts to limit thermally. This feels like a card that could really shine with more power (and under water) but obviously that won't matter if I brick it because of my ****ty soldering skills and/or not being prepared enough.

Regarding clearance issues: I've got the EK 3090 FE waterblock so I may have different issues (and it's possible that stacking would never work with this block so in hindsight I should have confirmed that first). I'll check on that when do I my water block assembly later this afternoon.
A stacked shunt only adds 0.5mm to the total height. Usually the shunts don't ever touch the backplate anyway so all you have to do is just apply kapton tape right behind the shunt and you're good. Don't overthink this. Just make sure the shunt is laid flat and level on top of the original shunt and there is no excess solder on top of the new shunt.

I linked the iron and the other items to get. The TS100 is the absolute minimum iron you should be using for shunts. Oldirdey, who is also friends with @dante`afk is the guy who recommended that iron. He's the same one from the 2080 Ti shunt video. Very experienced.
 

·
Registered
Joined
·
326 Posts
Great tips, thanks. I'll make sure to save this for later.

I think it doesn't help I bought a mediocre solder kit and me being a novice at soldering had oxidized my super skinny tip (which had terrible heat transfer at the tip but that's another issue) so I was attempting with a small but not tiny tip.

I think the extra motherboard is a great idea if I try this again. I've also got some extra graphics cards so I should definitely play around on those before trying this again.

Assuming I can revert the 2 resistors I soldered, clean up the short, and get everything working again, it may actually be better for me to first check performance of my card under the water block if nothing else as a baseline. I'm able to get 2070-2100 in games from this card before it starts to limit thermally. This feels like a card that could really shine with more power (and under water) but obviously that won't matter if I brick it because of my ****ty soldering skills and/or not being prepared enough.

Regarding clearance issues: I've got the EK 3090 FE waterblock so I may have different issues (and it's possible that stacking would never work with this block so in hindsight I should have confirmed that first). I'll check on that when do I my water block assembly later this afternoon.
i have a 3090FE shunted + EK wb, no issues with back plate, dont worry
 

·
Registered
Joined
·
94 Posts
Can someone check over my numbers please. The original resistors have all been stacked with 8moh. A couple of the rear ones need tidying up as I'm pretty sure it has caused the additional memory temps (I put 0.5 thermal pad on them to ensure they don't directly touch the backplate).

This is after running 3dmark timespy with power slider at max. The lines with the red marks have been timed by 1.63 which I believe is correct for this resistor arrangement (let me know if I'm wrong).

2489618


Temps aren't great but I'm using some cheap thermal paste and looking to water cool once I know it is all working good. Thanks.
 

·
Facepalm
Joined
·
9,883 Posts
Can someone check over my numbers please. The original resistors have all been stacked with 8moh. A couple of the rear ones need tidying up as I'm pretty sure it has caused the additional memory temps (I put 0.5 thermal pad on them to ensure they don't directly touch the backplate).

This is after running 3dmark timespy with power slider at max. The lines with the red marks have been timed by 1.63 which I believe is correct for this resistor arrangement (let me know if I'm wrong).

View attachment 2489618

Temps aren't great but I'm using some cheap thermal paste and looking to water cool once I know it is all working good. Thanks.
Is that on a FE card?
The 8 pins are perfect.
VRAM temps are not bad, but GPU Hotspot temp is terrible. 25C delta means you have a serious contact pressure problem with the core.
I guess it doesn't matter if you're getting rid of the stock cooler, but that shouldn't be left to run like that.
 

·
Registered
Joined
·
94 Posts
Is that on a FE card?
The 8 pins are perfect.
VRAM temps are not bad, but GPU Hotspot temp is terrible. 25C delta means you have a serious contact pressure problem with the core.
I guess it doesn't matter if you're getting rid of the stock cooler, but that shouldn't be left to run like that.
Thanks - Yeah FE.

So are the shunts working correctly? Shouldn't the max power usage be higher - Nearer 600w? GPU utilisation was max for most, few drops to 97%.
From my power reader at the mains I've seen a max of 670w while doing timesoy - I would have thought CPU / mobo power draw should be fairly low

VRAM is definitely hindered by the shunts. A couple of the rear felt a little proud of the adjacent chips. Even at old power levels, memory temps are a fair bit higher than they were and the heatsink isn't getting hot - Also I have a huge copper heatsink stuck to the backplate with a 120mm fan pulling heat away.

Regarding core temps - I think it could be the thermal pads. They are Arctic 1.5mm which are fairly firm (2mm on the rear which are obviously fine). I have replacement quality Gelids and Thermalright Odyssey for which I will use one of them. The thermal paste is some big tube of unknown unbranded stuff I primarily bought for testing stuff, rather than wasting quality stuff so that likely won't help.
 

·
Facepalm
Joined
·
9,883 Posts
Thanks - Yeah FE.

So are the shunts working correctly? Shouldn't the max power usage be higher - Nearer 600w? GPU utilisation was max for most, few drops to 97%.
From my power reader at the mains I've seen a max of 670w while doing timesoy - I would have thought CPU / mobo power draw should be fairly low

VRAM is definitely hindered by the shunts. A couple of the rear felt a little proud of the adjacent chips. Even at old power levels, memory temps are a fair bit higher than they were and the heatsink isn't getting hot - Also I have a huge copper heatsink stuck to the backplate with a 120mm fan pulling heat away.

Regarding core temps - I think it could be the thermal pads. They are Arctic 1.5mm which are fairly firm (2mm on the rear which are obviously fine). I have replacement quality Gelids and Thermalright Odyssey for which I will use one of them. The thermal paste is some big tube of unknown unbranded stuff I primarily bought for testing stuff, rather than wasting quality stuff so that likely won't help.
Yeah the Arctic pads are 1.5mm, shore 25 00, which is good enough. But even the worst paste on the market shouldn't cause a 25C core to core hotspot temp delta. Definitely check that out.

FE will not exceed about 580W even if you use 5 mOhm shunts, except in Quake 2 RTX and a few other games that are low on "Rasterization" usage and high on "Path tracing" usage. There are other power limits that can be seen in 'TDP Normalized' that the shunts do not control. You can see TDP Normalized in HWinfo (not in GPU-Z). TDP Normalized reports the highest normalized power rail of any rail, with respect to a "default" TDP for that rail, and a "maximum" TDP for that rail.

Run Timespy again and this time have the TDP and TDP Normalized rails visible in HWinfo64 (you may need to expand the values on the left side as the newest version has collapsible fields on the far left) and post a screenshot of the maximum values reported. If your main TDP (Sum of the 8 pins + PCIE Slot Power)is below 100% you're good. TDP Normalized will likely be much higher.

TDP Normalized will cause throttle at 100% TDP slider for any 'default' value for a rail which approaches that limit (starts slowly at about 95% Normalized) and 114% TDP (starts slowly at about 107% Normalized). This happens even if the board TDP draw is much lower (you can compare both TDP values in HWinfo64). Elmor thinks these limits are related to PLL, MSVDD or NVVDD power limits, so you can't do anything about that. HWinfo64 only reports the rails which NVAPI exposes, and these extra rails don't seem to be reported to anything except "Normalized."
It is known that the ROG Strix has higher internal power limits for these rails (e.g. Timespy Extreme with 5 mOhm stacked shunts on a ROG Strix [someone posted a screenshot of a completely unthrottled TSE run with just shunt mod only] will not trigger a power limit on the Strix, but will trigger TDP Normalized power throttling on the FE even at 114% slider).

Do you have a picture of the rear side of the card with the backplate removed? (clear, high resolution please?)

It's very important to make sure the shunts are flush. You can have extra solder around the "lower" side of the shunts, to make sure the contact between the two shunts is solid and complete (Remember to use 3M Kapton / Polyimide tape before doing this!!), but it is very important to have the top of the shunts solder free and flat, and the shunt to be flat against the lower shunt (not tilted like a see-saw) and not sitting on top of "extra solder" (that is what causes un-even level). Now you see why the Kapton tape is so important.

Also using ANY thermal pad on top of the stacked shunt is not the best choice. This will reduce thermal pad contact pressure with the VRAM itself and could even cause possible flexing. Just use a strip of "Kapton tape" above the shunt (don't put it on top of the shunt! Put it on the backplate, above the shunt, that way it will be easy to see if the shunt is pressing against it). As long as it's good quality like 3M, this will protect against shorts, after you trim the solder from the top of the shunt. Then you can do a quick screw-in of the backplate (torx and philips) then immediately unscrew it, then look at the Kapton tape for any rips or dents.
 

·
Registered
Joined
·
94 Posts
Run Timespy again and this time have the TDP and TDP Normalized rails visible in HWinfo64 (you may need to expand the values on the left side as the newest version has collapsible fields on the far left) and post a screenshot of the maximum values reported. If your main TDP (Sum of the 8 pins + PCIE Slot Power)is below 100% you're good. TDP Normalized will likely be much higher.
Thanks for all the information. Not sure I quite understand it all though!

I reset the information and ran just Timespy Extreme. Hit a higher power value. Are you talking about the 5th and 6th from bottom on this screenshot?

2489716


I won't run anymore though with these temps. Going to sort the rear shunts and see if I can spot any mounting problems on the front.

Here are some photos of the rear currently
2489717

2489718

2489720
2489719
2489721

2489722
 

·
Facepalm
Joined
·
9,883 Posts
Thanks for all the information. Not sure I quite understand it all though!

I reset the information and ran just Timespy Extreme. Hit a higher power value. Are you talking about the 5th and 6th from bottom on this screenshot?

View attachment 2489716

I won't run anymore though with these temps. Going to sort the rear shunts and see if I can spot any mounting problems on the front.

Here are some photos of the rear currently View attachment 2489717
View attachment 2489718
View attachment 2489720 View attachment 2489719 View attachment 2489721
View attachment 2489722
Yes TDP Normalized is what is throttling you. Without a bios editor there's no way you can avoid that. With the 114% TDP slider, normalized starts dropping your clocks when it reaches about 107%.

The first shunt looks okay, just remove that solder bump on the left side and smooth it out. Then remove the thermal pad above it and apply kapton tape on the backplate above it.
The second shunt looks more suspect. Not completely flat but it may still be ok. Try to flatten the right side a bit better then smooth out those solder bumps on top of it. Then apply a strip of kapton tape on the backplate above it.
The third shunt is messy. Flat but way too much solder on the top. Clean that off completely.

The key to the shunt level is to make 100% sure the top of the shunt's height DOES NOT exceed the height of the MLCC capacitors right next to it.
 

·
Registered
Joined
·
94 Posts
Bit hard to see but not exceeding the MLCC capacitors now. Thanks. Going to test it to see if memory temps are improved. If not then there must be something wrong with the reassembly on the front.
Next will be to plan installing the Bykski water block.
2489759

2489760

2489761

2489762
 
641 - 653 of 653 Posts
Top